CC-301i · Module 3
Cost Monitoring and Budgets
3 min read
Claude Code usage costs real money, and unmonitored usage leads to surprise bills. The token counter in Claude Code shows your current session usage. Track it — not obsessively, but enough to develop intuition for what costs what. Reading a 500-line file costs roughly 2,000 tokens. A typical coding session with moderate file reading and ten conversation turns costs 50,000-100,000 tokens. A heavy session with extensive file reading and refactoring can exceed 200,000 tokens.
The budgeting strategy is simple. Set a daily or weekly token budget based on your plan. Track actual usage against the budget. When usage consistently exceeds the budget, analyze where the tokens are going — usually it is unnecessary file reads, verbose prompts, or sessions that run too long without /clear. Optimize the highest-cost sessions first. Often a single habit change — reading specific functions instead of entire files, or starting fresh sessions for independent tasks — reduces token usage by 30-40% with no reduction in output quality.
Do This
- Track token usage weekly and compare against your budget
- Identify high-cost sessions and analyze what drove the usage
- Read specific functions or sections, not entire files when possible
- Start fresh sessions for independent tasks instead of reusing stale ones
Avoid This
- Ignore token usage until the bill arrives
- Optimize every prompt to save 50 tokens — the ROI is too low
- Avoid using Claude to save tokens — the time savings justify the cost
- Run one session all day — context degradation increases corrections which increases cost